The present invention generally relates to parallel image processing system and more particularly to a recognition of image using such a parallel image processing system.
Conventionally, there are various systems proposed for detecting contrasts, isolated points, direction of lines, end points and the like by simulating the function of the optic nerve system of a biological body. Among others, Japanese Patent Publication No. 50-34901 and Japanese Patent Publication No. 53-23987 describe a parallel image processing system performing such a function. However, no system is reported so far which can detect the movement of object by simulating the optical nerve system of biological body.
In a conventional image input system used in a robot and the like, an input image is sampled after picturing by a television camera. The image thus sampled is then digitized and processed time sequentially by a digital computer. Such a conventional procedure has a problem in that it takes a substantial time for sampling, digitization, feature extraction, and the like. In order to solve this problem, there is proposed a use of two dimensional array of photodetectors for receiving the input image. In this system, feature extraction is performed by combining output signals of the various photodetectors. In such a system, however, there is a problem in that electrical wiring for connection of the photodetectors becomes complex.
In the aforementioned conventional image processing system for detecting a two dimensional image based on digital processing, the detection of slope of a line image, which is essential for recognition of the image, is performed by a number of image processing steps such as preliminary processing, feature extraction, pattern matching and the like. Such a processing is performed time sequentially on the basis of two dimensional image data stored in an image memory. Thus, the procedure is time consuming and needs a large memory space.